CLAY: a generative AI model for earth
, , , , , , , , ,

CLAY: A generative AI model for Earth

The democratization of geospatial data and services could give rise to cutting-edge and innovative strategies. Data and services available to everyone and anyone for utilization are essential. For some time, governments, communities, and organizations have made geospatial data and services freely available to the general public.

Furthermore, this method promotes a deeper knowledge of the systems, leading to collaborative ways to mitigate worldwide challenges. One such impending global threat has been manifesting itself more often than usual. That is ‘Climate Change’.

It has not arisen overnight, nor can it be solved overnight. But it can be tackled by smart solutions collaboratively. There is one such big effort by ‘Radiant Earth’. It is called ‘CLAY’.

What is CLAY?

CLAY is a mission. CLAY is a joint venture to make a difference. It is a fiscally sponsored project of the non-profit ‘Radiant Earth’. Let me phrase it this way. It is a thriving, sophisticated system. 

Let me put it this way. Geospatial data can be structured via the dynamic CLAY system for practical applications. Users can enhance and freely customize this process for the project using Clay’s app and third-party apps.

 It is an open-source AI model or rather ChatGPT for Earth. We mold our future the way we want. The Clay model can be utilized in three major ways:

  1. Create semantic embeddings in any location and time.
  2. Fine-tune the model for future tasks including classification, regression, and generative tasks.
  3. Use the model as the foundation for additional models.

What is CLAY’s main ingredient?

CLAY is a foundational model that utilizes ‘Vision Transformer’. Vision Transformer is a kind of architecture that is adapted to understand geospatial and temporal relations on Earth Observation data. The model is trained via Self-supervised learning (SSL) using a Masked Autoencoder (MAE) method.

  • Clay leverages AI, satellite pictures, and other geographical data to categorize information on what is happening in specific locations worldwide.
  • This sophisticated model is fed millions of satellite photographs and imagery. Then, AI technologies, study these data while learning about our continuously changing Earth through those images. 
  • As it learns, we gauge how those skills increase its ability to perform critical tasks such as making land cover maps, detecting crops or burn scars, and estimating carbon stock.

What is a Vision Transformer?

Vision Transformer (ViT) is a cutting-edge neural network design that transforms how we detect and understand images. The Vision Transformer (ViT) paradigm was initially outlined in a 2021 conference research report.


ViT pioneered an unconventional way of image analysis that divides images into fragments and takes advantage of self-attention mechanisms. This allows the model to incorporate both local and global linkages into images. ViT’s core layer, encoder layers, are made up of multi-head self-attention and feedforward neural networks.

  • The multi-head self-attention system detects the link between specific patches in the input sequence. This is how it prioritizes essential patches.


  • Following self-attention, the output of each patch’s self-attention mechanism is routed through a feedforward neural network. 
  • Introducing a non-linear nature and allowing the model to adjust its comprehension of intricate interactions between patches is the aim of the feedforward network.

Therefore, ViT performs exceptionally well on a variety of computer vision tasks.

What problems CLAY can fix?

It is imperative to raise awareness about climate change. Allowing the curve to bend takes a team effort, particularly when working with high-quality and high-integrity earth system data.

As an open-source technology, CLAY is specially made to let organizations produce and use earth intelligence that supports groups working to identify persistent solutions to the repercussions of climate change. In addition, it offers roadways such as assessment tools, apps, APIs, advanced models, and filtered raw data analysis.

To be honest, we don’t have nearly as much earth data at our fingertips as would be demanded to put forward substantial solutions. Additionally, CLAY promises to encourage more communities, climate agencies, institutions, and organizations to work together to help close this gap. 

CLAY data reliability

Erroneous data can lead to a misunderstanding of the patterns. When such data is loaded into an AI model, it can result in inaccurate predictions. We don’t need decisions made based on data that is not accurate. CLAY relies on data from trusted sources. NASA and the European Space Agency provide trusted open-source data.